EUVIMEDEUVIMED entdecken
Crossref

Convolutional-Neural-Network-Based Identification of Leak-Related Acoustic Changes Using Multi-day Sensor Data in Water Distribution Systems

Ruilin Liu, Chi Zhang, Mark Stephens, Nhu Cuong Do, Wei Zeng, Martin Lambert

Civil Engineering Sciences · 2026

DOI: 10.34133/cesci.0035 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Civil Engineering Sciences, 2026. Verfasst von Ruilin Liu und 5 weiteren Autor:innen. Die Arbeit nennt 36 Literaturangaben. Gefördert von: Australian Research Council.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

The timely identification of leak-related acoustic changes in operational water distribution systems is essential for leakage management, as it enables utilities to recognize developing faults before they progress into major pipe failures. However, this task remains challenging because long-term acoustic signals are affected by environmental noise, sensor heterogeneity, and temporally evolving fault-related behaviors. This study presents an artificial-intelligence-driven framework for long-term acoustic monitoring in smart water infrastructure using a compact composite multi-day power spectrum (CMPS) representation and a squeeze-and-excitation convolutional neural network (SE-CNN). By aggregating acoustic information across consecutive days, the CMPS emphasizes temporally persistent spectral behaviors associated with developing faults while suppressing transient environmental disturbances. The SE-CNN further enhances feature utilization through adaptive channel-wise attention, enabling the robust identification of informative spectral structures under noisy and heterogeneous operational conditions. The proposed framework was evaluated using real-world acoustic data collected from the Adelaide smart water network and demonstrated balanced classification performance across 4 acoustic-state classes (normal, states, banded, and change), achieving F 1 -scores of 0.95, 0.96, 0.94, and 0.90, respectively. A controlled field experiment using a co-located accelerometer and hydrophone provided preliminary evidence that the persistence-based representation can transfer across these 2 sensing modalities under the tested pipe and leak condition without sensor-specific retraining. The results demonstrate that persistence-aware spectral representation combined with attention-guided deep learning provides an effective and computationally efficient approach for identifying leak-related acoustic changes and supporting intelligent infrastructure monitoring in smart water networks.

Förderung

  • Australian Research Council

Zitieren

APA

Liu, R., Zhang, C., Stephens, M., Do, N. C., Zeng, W., & Lambert, M. (2026). Convolutional-Neural-Network-Based Identification of Leak-Related Acoustic Changes Using Multi-day Sensor Data in Water Distribution Systems. Civil Engineering Sciences. https://doi.org/10.34133/cesci.0035

Vancouver

Liu R, Zhang C, Stephens M, Do NC, Zeng W, Lambert M. Convolutional-Neural-Network-Based Identification of Leak-Related Acoustic Changes Using Multi-day Sensor Data in Water Distribution Systems. Civil Engineering Sciences. 2026. doi:10.34133/cesci.0035

BibTeX
@article{Liu2026,
  author = {Liu, Ruilin and Zhang, Chi and Stephens, Mark and Do, Nhu Cuong and Zeng, Wei and Lambert, Martin},
  title = {Convolutional-Neural-Network-Based Identification of Leak-Related Acoustic Changes Using Multi-day Sensor Data in Water Distribution Systems},
  journal = {Civil Engineering Sciences},
  year = {2026},
  doi = {10.34133/cesci.0035}
}
Kostenlos starten →